Transcript Review of Linear Algebra - Carnegie Mellon University

+
Review of Linear Algebra
10-725 - Optimization
1/14/10 Recitation
Sivaraman Balakrishnan
+
Outline
 Matrix
subspaces
 Linear
independence and bases
 Gaussian
 Eigen
elimination
values and Eigen vectors
 Definiteness
 Matlab
essentials
 Geoff’s LP sketcher
 linprog
 Debugging and using documentation
Basic concepts

Vector in Rn is an ordered
set of n real numbers.




1
 
6
 3
 
 4
 
e.g. v = (1,6,3,4) is in R4
A column vector:
A row vector:
m-by-n matrix is an object
with m rows and n columns,
each entry filled with a real
(typically) number:
1
6 3 4
1 2 8


 4 78 6 
9 3 2


Basic concepts - II
 Vector
dot product:
 Matrix
product:
u  v  u1 u2   v1 v2   u1v1  u2v2
 a11 a12 
 b11 b12 
, B  

A  
 a21 a22 
 b21 b22 
 a11b11  a12b21 a11b12  a12b22 

AB  
 a21b11  a22b21 a21b12  a22b22 
+
Matrix subspaces
 What

is a matrix?
Geometric notion – a matrix is an object that “transforms” a
vector from its row space to its column space
 Vector
space – set of vectors closed under
scalar multiplication and addition
 Subspace
– subset of a vector space also
closed under these operations

Always contains the zero vector (trivial subspace)
+
Row space of a matrix

Vector space spanned by rows of matrix

Span – set of all linear combinations of a set of vectors

This isn’t always Rn – example !!

Dimension of the row space – number of linearly
independent rows (rank)

We’ll discuss how to calculate the rank in a couple of slides
+
Null space, column space

Null space – it is the orthogonal compliment of the row space

Every vector in this space is a solution to the equation

Ax = 0

Rank – nullity theorem

Column space

Compliment of rank-nullity
+
Linear independence

A set of vectors is linearly independent if none of them can
be written as a linear combination of the others

Given a vector space, we can find a set of linearly
independent vectors that spans this space

The cardinality of this set is the dimension of the vector
space
+
Gaussian elimination

Finding rank and row echelon form

Applications

Solving a linear system of equations (we saw this in class)

Finding inverse of a matrix
+
Basis of a vector space

What is a basis?
 A basis is a maximal set of linearly independent vectors and a
minimal set of spanning vectors of a vector space

Orthonormal basis
 Two vectors are orthonormal if their dot product is 0, and each
vector has length 1
 An orthonormal basis consists of orthonormal vectors.

What is special about orthonormal bases?
 Projection is easy
 Very useful length property
 Universal (Gram Schmidt) given any basis can find an
orthonormal basis that has the same span
+
Matrices as constraints

Geoff introduced writing an LP with a constraint matrix

We know how to write any LP in standard form

Why not just solve it to find “opt”?
A special basis for square matrices

The eigenvectors of a matrix are unit vectors that satisfy
 Ax = λx

Example calculation on next slide

Eigenvectors are orthonormal and eigenvalues are real for
symmetric matrices

This is the most useful orthonormal basis with many
interesting properties
 Optimal matrix approximation (PCA/SVD)

Other famous ones are the Fourier basis and wavelet basis
Eigenvalues
(A – λI)x = 0
λ is an eigenvalue iff det(A – λI) = 0
Example:
4
5 
1


A  0 3 / 4 6 
0
0 1 / 2 

4
5 
1  


det(A  I )   0
3/ 4  
6   (1   )(3 / 4   )(1 / 2   )
 0
0
1 / 2   

  1,   3 / 4,   1 / 2
+
Projections (vector)
(2,2,2)
b = (2,2)
(0,0,1)
(0,1,0)
(1,0,0)
 2   1 0 0  2 
  
 
 2    0 1 0  2 
 0   0 0 0  2 
  
 
a = (1,0)
 2
ab
c  T a   
aa
 0
T
+

Matrix projection
Generalize formula from the previous slide


Special case of orthonormal matrix


Projected vector = (QTQ)-1 QTv
Projected vector = QTv
You’ve probably seen something very similar in least squares regression
Definiteness

Characterization based on eigen values

Positive definite matrices are a special sub-class of invertible
matrices

One way to test for positive definiteness is by showing


xTAx > 0 for all x
A very useful example that you’ll see a lot in this class

Covariance matrix
Matlab Tutorial - 1

Linsolve


Stability and condition number
Geoff’s sketching code – might be very useful for HW1 
Matlab Tutorial - 2

Linprog – Also, very useful for HW1 

Also, covered debugging basics and using Matlab help
+
Extra stuff

Vector and matrix norms

Matrix norms - operator norm, Frobenius norm

Vector norms - Lp norms

Determinants

SVD/PCA